LibreFang Changelog

LibreFang 2026.4.28

2026-04-28T00:00:00Z

] blocks via Accept-Language (#3424) (@houko) - Regroup metrics, surface unused per-agent data, collapse endpoints (#3427) (@houko) - Click anywhere on a channel card to open the drawer (#3434) (@houko) - Rich markdown help drawer + page coverage + UserBudget redesign (#3435) (@houko) ### Fixed - Unbreak main — namespace traversal substring + openapi.json bump (#3258) (@houko) - Add dbus to buildInputs to fix failing build (#3263) (@FrantaNautilus) - Install libdbus-1 so image builds and starts (closes #3259) (#3265) (@houko) - Keyring is target-conditional so musl/android cross builds compile (#3267) (@houko) - Copy deploy/ into builder so include_str! observability assets resolve (closes #3259) (#3268) (@houko) - Show declared tools in editor and persist to **disk** (#3269) (@leszek3737) - Recognize BYTEPLUS_API_KEY in provider key checks (#3274) (@houko) - Silence three sources of routine WARN log spam (#3275) (@houko) - Skip OTLP exporter when no collector is reachable (#3276) (@houko) - Point at recovery commands when boot integrity check fails (#3277) (@houko) - Align model_catalog/routing tests with current registry (#3280) (@houko) - Refresh provider list after Test button so latency shows (#3288) (@houko) - Wire missing applyDatePreset for quick-pick buttons (#3289) (@houko) - Align useDeleteWorkflow test with removeQueries semantics (#3290) (@houko) - Use correct path + auth for Anthropic-protocol providers (#3292) (@houko) - Add missing librefang-llm-drivers dep to unbreak main (#3294) (@houko) - Stop bypassing needs-changes via comment inference / push (#3312) (@houko) - Treat Anthropic 401/403 as reachable, not auth-failed (#3316) (@houko) - Decouple model-id assertions from registry catalog state (#3317) (@houko) - Enforce deterministic ordering for LLM-bound registries (#3325) (@houko) - Install libdbus-1-dev for glibc Linux CLI builds (#3357) (@houko) - Drop layout/AnimatePresence from StaggerList to unblock clicks (#3415) (@houko) - Regenerate kernel config schema golden after thread-ownership field (#3417) (@houko) - Drawer not opening on hands page (DrawerPanel mount race) (#3421) (@houko) - Add /api/auto-dream/status to dashboard read allowlist (#3426) (@houko) - Scale Top Endpoints status bar with call volume (#3428) (@houko) - Exempt loopback + cheaper cost for dashboard polls (#3430) (@houko) ### Changed - Tidy env_passthrough nits from #3219 review (#3273) (@houko)

Documentation, maintenance, and other internal changes

LibreFang 2026.4.27

2026-04-27T00:00:00Z

### Added - TUI setup wizard now offers `microsoft`, `zai`, `zai_coding`, `volcengine`, `volcengine_coding`, `byteplus`, `byteplus_coding` alongside the existing first-run options. The wizard's PROVIDERS list had drifted from `PROVIDER_REGISTRY` and silently hid these from new installs; a unit test now pins these entries against future regressions. (@houko) - Treat CLI logins as first-class default providers (#3061) (@houko) - Grafana Tempo + business-level span instrumentation (#3064) (@houko) - /new creates a new session instead of resetting the current one (#3071) (@neo-wanderer) - Support image-generation models (registry modality field) (#3074) (@houko) - Wire chat attachment uploads in ChatPage (#3075) (@houko) - Add Novita AI as OpenAI-compatible provider (#3076) (@houko) - Agent name prefix on outbound + Signal plain-text default (#3077) (@houko) - SSE attach endpoint for multi-client session co-watching (#3078) (@houko) - Add SearXNG self-hosted search provider (#3079) (@houko) - Add AWS Bedrock provider with Bearer token auth (#3080) (@houko) - AuditCheck framework + first 3 CLAUDE.md gotcha checks (#3082) (@houko) - Add LlmFamily enum + LlmDriver::family() (#3083) (@houko) - SSE attach hook for multi-client session co-watching (#3087) (@houko) - Add ToolApprovalClass + tool_classifier (no behavior change yet) (#3092) (@houko) - Session lifecycle event bus (additive, no subscribers yet) (#3093) (@houko) - Support PDF and text/code file attachments end-to-end (#3094) (@houko) - Trajectory export endpoint with privacy redaction (#3097) (@houko) - Extend detect_embedding_provider with vLLM + LM Studio fallback (#3099) (@houko) - Cron multi-destination delivery with failure isolation (#3102) (@houko) - UI for cron multi-destination delivery targets (#3103) (@houko) - Cache /config + reject pageno=0 + annotate truncation (#3108) (@houko) - Re-read agent context.md per turn (#3115) (@houko) - Central slash command registry (PR-1/3) (#3122) (@houko) - Slash command registry — CLI/TUI surface (PR-2/3) (#3123) (@houko) - Configurable max history messages (per-agent + global override) (#3125) (@neo-wanderer) - System_and_3 prompt cache stamping for Anthropic (M1) (#3126) (@houko) - ParallelSafety projection for batch tool dispatch (PR-1/6) (#3127) (@houko) - Plan_batch + path-overlap planner for tool dispatch (PR-2/6) (#3129) (@houko) - Model metadata lookup pipeline (PR-1/3, layers 1+2+5) (#3133) (@houko) - Model metadata L3 cache + L4 Ollama probe (PR-2/3) (#3134) (@houko) - Model metadata L4 Anthropic + OpenAI-compat probes (PR-2.5/3) (#3140) (@houko) - KernelConfig.parallel_tools section (PR-3/6) (#3144) (@houko) - Cron pre_script + silent_marker schema (PR-1/3) (#3145) (@houko) - Cache_hit_ratio metric + trajectory field (M2/2) (#3149) (@houko) - Agent detail drawer + filter pill i18n (#3159) (@houko) - Right-side drawer pattern for inspect-detail surfaces (#3166) (@houko) - Convert hand detail panel to drawer variant (#3168) (@houko) - Roll out drawer/panel pattern across all page modals (#3175) (@houko) - Add Jaeger as second trace backend alongside Tempo (#3176) (@houko) - Granular MCP taint policy + dashboard tree editor (closes #3050) (#3193) (@houko) - Jaeger trace backend + Loki/Alloy logs + CLI wiring (#3194) (@houko) - Per-(agent, session) liveness tracking and session-scoped stop (#3195) (@houko) - RBAC M2 — audit user/channel attribution + stable UserId (#3054) (#3196) (@houko) - Hot-reload log_level via dashboard without daemon restart (#3200) (@houko) - RBAC M4 — channel-native role mapping (Telegram/Discord/Slack) (#3054) (#3202) (@houko) - RBAC M5 — audit query/export + per-user budget API (#3054) (#3203) (@houko) - RBAC M3 — per-user tool policy + memory namespace ACL (#3054) (#3205) (@houko) - RBAC M6 — dashboard (users, identity linking, simulator, CSV import + stubs) (#3054) (#3209) (@houko) - Per-agent + global lane caps for trigger dispatch (#3210) (@neo-wanderer) - Auto-download voice messages mirroring file path (#3212) (@neo-wanderer) - Wip (#3213) (@houko) - Hand agent runtime overrides with restart persistence (#3216) (@leszek3737) - Deliver HealthCheckFailed to notification.alert_channels (#3218) (@neo-wanderer) - Per-user budget write/clear endpoints + dashboard editor (#3224) (@houko) - Activate AuditPage now that M5 audit endpoints shipped (#3225) (@houko) - Per-action retention policy with chain-anchor trim (#3227) (@houko) - RBAC effective-permissions snapshot — wire simulator (#3054) (#3228) (@houko) - RBAC M3 — per-user policy GET/PUT + dashboard editor (#3229) (@houko) - RBAC — single-decision authz/check endpoint (#3054) (#3231) (@houko) - User-list summary flags + custom channel rule editor (#3229 follow-up) (#3232) (@houko) - Owner-only API key rotation with live session kill (#3233) (@houko) - External mount points in agent.toml (#3230) (#3234) (@houko) - Channel field as dynamic dropdown with custom fallback (#3248) (@houko) - URL-synced filters, JSON export, row detail modal (#3252) (@houko) - Move filters into right-docked drawer (#3254) (@houko) - BeforePromptBuild hook can contribute labeled DynamicSection injected into the system prompt, with 8KiB per-section / 32KiB total caps (closes #3326) (#3358) (@houko) ### Fixed - Reconnect WhatsApp gateway after transient disconnects (#21) (@houko) - Render connection screen via custom URI scheme (closes #3052) (#3056) (@houko) - Create log dir + open log before stdout redirect (#3057) (@houko) - Surface CLI logins as their own providers, not API-provider fallbacks (#3059) (@houko) - Pre-create logs dir in entrypoint (defense for #3058) (#3060) (@houko) - Bundle compose stack in-binary, add OTLP collector (#3062) (@houko) - Create HTTP trace spans at INFO so OTel exporter sees them (#3063) (@houko) - Move env_filter to fmt layer so OTel sees INFO spans (#3065) (@houko) - Drop ingester/compactor from Tempo config (#3067) (@houko) - Boot-time TOML drift detection now reaches hand agents (#3068) (@neo-wanderer) - Reprobe local providers every 60s + refresh on test (#3069) (@houko) - Add missing files to src to fix librefang-cli build (#3073) (@FrantaNautilus) - Honor session_mode=new with per-fire isolated sessions (#3081) (@houko) - Copilot streaming empty tool calls + Claude assistant strip (#3084) (@houko) - Gemini array-items default + first-message-must-be-user (#3085) (@houko) - Safe UTF-8 boundary in three remaining truncation sites (#3086) (@houko) - PowerShell sandbox bypass + agent-config persistence + WS race + Revolt self-host (#3088) (@houko) - Cron preservation across hand reactivation + telegram startup timeout + token estimation includes ToolUse (#3090) (@houko) - Capture text from intermediate tool_use iterations (#3091) (@houko) - Percent-decode WS auth token to preserve base64 characters (#3095) (@houko) - Skip heartbeat timeout for agents in their idle grace window (#3096) (@houko) - Handle BrokenPipe gracefully in doctor --json (#3100) (@houko) - UTF-8-safe error truncation + 502/504 retry + response classify tests (#3104) (@houko) - Cap accumulated_text + document streaming non-redelivery contract (#3106) (@houko) - Cron dedupe + next_run + token_length annotation (#3109) (@houko) - Sticky has_processed_message replaces time-based grace (#3111) (@houko) - Use 127.0.0.1 instead of localhost for local LLM URLs (#3112) (@houko) - Pass agents_dir to hand route candidate scan to silence WARN flood (#3113) (@houko) - Close non-loopback auth bypass when api_key is empty (#3114) (@houko) - Downgrade pure-normalization to debug, keep WARN for real repair (#3117) (@houko) - Use "default" provider/model in custom-agent template (#3121) (@houko) - Forward api_key as Bearer in local provider probe (#3128) (@houko) - Degrade Memory page gracefully when proactive memory is disabled (#3131) (@houko) - Allow named workspaces in read-side path resolution (#3137) (@neo-wanderer) - Unbreak cron_delivery tests + move guards to input validation (#3139) (@houko) - Unbreak local provider config in GUI (#3141) (@houko) - Re-render hand [[settings]] tail after boot-time TOML drift (#3142) (@neo-wanderer) - Relax probe timeout for remote local-provider URLs (#3146) (@houko) - Preserve tool annotations for parallel safety classification (PR-6/6) (#3147) (@houko) - Include SearXNG in web_search_available check (#3152) (@houko) - Drop redundant runtime SSRF check in deliver_webhook (#3155) (@houko) - Add .desktop entry and install icon (#3157) (@FrantaNautilus) - Seed [[settings]] defaults into hand instance config on activation (#3160) (@houko) - Skip empty Blocks when stamping prompt cache markers (review fix for #3126) (#3161) (@houko) - Expose vLLM + LM Studio in embedding provider dropdown (refs #3138) (#3162) (@houko) - Re-render Reference Knowledge + Your Team tails after TOML drift (#3164) (@houko) - Provide .desktop entry and icon for librefang-desktop (#3156) (#3165) (@houko) - Regenerate config_schema golden after parallel_tools addition (#3167) (@houko) - Stop drawer scroll chaining into the page (#3169) (@houko) - Observability auto-start opt-in + home_dir isolation + RAII cleanup (#3170) (@houko) - Surface provider model list above the fold (#3179) (@houko) - Wire OS keyring (libsecret/Keychain/Credential Manager) (#3180) (@houko) - Wrap with wrapGAppsHook3 so tray icon resolves on NixOS (#3197) (@houko) - Probe OpenAI fallback for ollama-slot servers, hide non-discovered local models (#3204) (@houko) - Correct max_level_hint test assertions (#3206) (@houko) - Correct max_level_hint test assertions (#3207) (@houko) - Set sender_user_id metadata so RBAC works in groups (#3215) (@neo-wanderer) - Serialize channel config writes via toml_edit + lock (#3183) (#3223) (@houko) - Attribute loopback callers to user_api_keys when token provided (#3236) (@houko) - Invalidate effective-permissions on policy/budget mutations (#3228 follow-up) (#3237) (@houko) - Prefix sender_chat ids so they can't collide with user namespace (#3215 follow-up) (#3238) (@houko) - RBAC M3 follow-up — memory ACL fail-closed for anonymous callers (#3239) (@houko) - Include prev_hash so verifiers can replay the chain (#3203 follow-up) (#3240) (@houko) - RBAC M4 follow-up — role_cache reload + Telegram DM owner-escalation (#3241) (@houko) - Mark scope as user_policy_only to match implementation (#3231 follow-up) (#3242) (@houko) - Attribute admin actions to caller + log old->new diffs (#21 follow-up) (#3245) (@houko) - Harden CSV import + flag identity-link risk (#3209 follow-up) (#3246) (@houko) - RBAC M3 follow-up — namespace traversal + case-insensitive deny + memory audit emit (#3205) (#3249) (@houko) - Autonomous-loop tool calls bypass user gate (closes #3243) (#3251) (@houko) - Channel dropdown uses /api/channels for full 44-adapter list (#3253) (@houko) - Enforce deterministic ordering for LLM-bound MCP server / skill registries to stabilize provider prompt cache (closes #3298) (#3325) (@houko) ### Changed - Derive JSON Schema from KernelConfig via schemars (#3055) (@houko) - Extract SessionStore trait alongside SQLite substrate (#3089) (@houko) - Make bridge helpers crate-private (#3181) (@houko) - Remove unused public helpers (#3182) (@houko) - Tighten visibility of internal request structs (#3184) (@houko) - Merge duplicate type definitions across crates (#3185) (@houko) - Rename Action enums to disambiguate from domain types (#3188) (@houko) - **BREAKING**: Split coding-provider API keys onto dedicated env vars — `byteplus_coding` now reads `BYTEPLUS_CODING_API_KEY` (was `BYTEPLUS_API_KEY`), `volcengine_coding` reads `VOLCENGINE_CODING_API_KEY` (was `VOLCENGINE_API_KEY`), `zai_coding` reads `ZAI_CODING_API_KEY` (was `ZHIPU_API_KEY`), `zhipu_coding` reads `ZHIPU_CODING_API_KEY` (was `ZHIPU_API_KEY`). Per-token siblings (`byteplus`, `volcengine`, `zai`, `zhipu`) keep their original env vars. Set the new env var if you use any `_coding` provider. (#3279) (@houko) - **BREAKING**: Register `microsoft` (GitHub Models / Azure AI Inference) as an explicit driver-registry entry with its own `GITHUB_MODELS_TOKEN` env var, distinct from `github-copilot`'s `GITHUB_TOKEN`. Same PAT works for both, but the env vars are now separate so configuring one product no longer auto-activates the other in the model picker. Set `GITHUB_MODELS_TOKEN` if you use the `microsoft` provider. (#3281) (@houko) - **BREAKING**: Split `zai` from sharing `ZHIPU_API_KEY` with `zhipu` — `zai` (api.z.ai) now reads `ZAI_API_KEY` while `zhipu` (open.bigmodel.cn) keeps `ZHIPU_API_KEY`. Same Zhipu credential value works for both, but the env vars are now separate so configuring one no longer auto-activates the other. Set `ZAI_API_KEY` if you use the `zai` provider. (#3285) (@houko) ### Documentation - Add tool_timeouts configuration documentation (#3098) (@leszek3737) - Backfill reference for cron / config / providers / channels / api / observability (#3189) (@houko) - Clarify worktree continuation drives to PR (#3190) (@houko) - Align left nav with file tree (#3199) (@houko) - Backfill source-vs-doc gaps (providers / channels — config / API / CLI to follow) (#3201) (@houko) - Drop HTML comment that broke Deploy Docs on main (#3208) (@houko) - Align Chinese translations with English source (#3220) (@houko) ### Maintenance - Rename normalize_schema_recursive + warn on items fallback (#3105) (@houko) - Document apply_agent_prefix idempotency caveats (#3107) (@houko) - Timing-side-channel mitigation in percent_decode (#3110) (@houko) - Align localhost test expectations with #3112 default change (#3118) (@houko) - Ignore local .plans/ working notes directory (#3130) (@houko) - Sync librefang-types tracing dep into Cargo.lock (#3132) (@houko) - Unbreak main — cargo fmt for model_metadata.rs (#3150) (@houko) - Unbreak main — fix clippy manual_pattern_char_comparison (#3153) (@houko) - Hand-level skills propagation regression for #3135 (#3163) (@houko) - Pull librefang-api into selective lane on librefang-types changes (#3171) (@houko) - Drop LEGACY_TEAM_TAIL_MARKER fallback (#3177) (@houko) - Install libdbus-1-dev for OpenAPI Drift job (#3186) (@houko) - Remove unused dependencies across workspace (#3187) (@houko) - Pin push_notification routing for health_check_failed (#3222) (@houko) - Unbreak typecheck on sessions-stream test (#3235) (@houko) - Unbreak typecheck on UserBudgetPage + duplicate type export (#3244) (@houko) ### Other - Unbreak main — use local user_api_keys snapshot (#3250) (@houko)

LibreFang 2026.4.24

2026-04-24T00:00:00Z

### Added - Per-tool timeout overrides via [tool_timeouts] (#2990) (@houko) - Attach to remote CDP endpoint instead of spawning Chromium (#2991) (@houko) - Attach to remote CDP endpoint instead of spawning Chromium (#2993) (@houko) - Configurable cron session size limit (#2994) (@houko) - REST API for task_queue + max_retries TTL enforcement (#2997) (@houko) - Generic OpenAI-compat driver for user-defined image providers (#2998) (@houko) - Per-tool / per-path taint policy with TaintRuleId skip API (#2999) (@houko) - Per-tab session_id on WebSocket + URL-driven ChatPage (incremental on #2989) (#3001) (@neo-wanderer) - Vacuum sqlite after session prune at startup (#3002) (@houko) - Add TransformToolResult hook for plugin tool-result rewriting (#3003) (@houko) - Add per-provider request_timeout_secs config (#3004) (@houko) - Preserve @mention context and show reaction processing state (#3005) (@houko) - Write compaction summaries in the user's conversation language (#3007) (@houko) - Add media attachment delivery support (#3008) (@houko) - Add reactions_enabled toggle for processing state indicators (#3009) (@houko) - Add wakeAgent gate for cron script pre-check (#3010) (@houko) - Add deliver_only mode for zero-LLM push notifications (#3011) (@houko) - Add send_voice and dm/group message policies (#3012) (@houko) - Per-agent ChannelOverrides in AgentManifest (#3020) (@DaBlitzStein) - Tee foreground daemon logs to timestamped daily files (#3022) (@houko) - Add POST /api/tools/{name}/invoke for direct tool execution (#3025) (@houko) - Auto-generate Python/JS/Go/Rust SDKs from openapi.json (#3046) (@houko) - Lazy tool loading via tool_load/tool_search (closes #3044) (#3047) (@houko) ### Fixed - Resolve 2937, build of both librefang-cli and librefang-desktop on NixOS (#2974) (@FrantaNautilus) - Infer Ollama model capabilities from families metadata (#2987) (@houko) - Include stdio server arg paths in MCP roots capability (#2988) (@houko) - Per-request session_id override on message send (#2989) (@houko) - Inject bot aliases into reply_precheck classifier prompt (#2992) (@houko) - Tolerate trailing reasoning tokens in tool call arguments (#2995) (@houko) - Detect vision/embedding capabilities for Ollama local models (#2996) (@houko) - Fix connection screen IPC on Windows + add uninstall button (#3000) (@houko) - Restore audit polling to 30s, drop expensive verify refetchInterval (#3006) (@houko) - Add missing task_get and task_update_status to stub KernelHandle impls (#3013) (@houko) - Guard max_tokens against zero to prevent HTTP 400 (#3014) (@houko) - Retry LLM stream on transient errors and add SSL/TLS error patterns (#3015) (@houko) - Detect macOS Chrome .app bundle for browser hand (#3021) (@houko) - Gate foreground tee behind #[cfg(unix)]; fix clippy warnings (#3024) (@houko) - Cascade parent /stop into agent_send subagents (#3044 follow-up) (#3048) (@houko) - Add plaintext fallback when editMessageText HTML is rejected (#3051) (@DaBlitzStein) ### Changed - Add QueryOverrides support, use withOverrides consistently (#2981) (@leszek3737) ### Performance - Optimize React components (#2979) (@leszek3737) - Narrow mutation cache invalidation and fix missing invalidations (#2980) (@leszek3737) ### Maintenance - Remove deprecated providers ai21, aider, chutes, venice (#3023) (@houko) - Bump actions/cache from 4 to 5 (#3026) (@app/dependabot) - Bump rustls from 0.23.37 to 0.23.39 (#3027) (@app/dependabot) - Bump webpki-roots from 1.0.6 to 1.0.7 (#3028) (@app/dependabot) - Bump tokio from 1.50.0 to 1.52.1 (#3029) (@app/dependabot) - Bump cbc from 0.1.2 to 0.2.0 (#3030) (@app/dependabot) - Bump aes from 0.8.4 to 0.9.0 (#3031) (@app/dependabot) - Bump tauri-plugin-dialog from 2.6.0 to 2.7.0 (#3032) (@app/dependabot) - Bump semver from 1.0.27 to 1.0.28 (#3033) (@app/dependabot) - Bump rmcp from 1.3.0 to 1.5.0 (#3034) (@app/dependabot) - Bump tauri-plugin-single-instance from 2.4.0 to 2.4.1 (#3035) (@app/dependabot) - Bump wasmtime from 43.0.1 to 44.0.0 (#3036) (@app/dependabot) - Bump open from 5.3.3 to 5.3.4 (#3037) (@app/dependabot) - Bump rustix from 0.38.44 to 1.1.4 (#3038) (@app/dependabot) - Bump lettre from 0.11.20 to 0.11.21 (#3039) (@app/dependabot) - Bump uuid from 1.23.0 to 1.23.1 (#3040) (@app/dependabot) - Bump rustls-connector from 0.22.0 to 0.23.0 (#3041) (@app/dependabot) - Bump axum from 0.8.8 to 0.8.9 (#3042) (@app/dependabot) - Bump seccompiler from 0.4.0 to 0.5.0 (#3043) (@app/dependabot)

LibreFang 2026.4.23

2026-04-23T00:00:00Z

### Added - Auto-reset stuck in_progress tasks after TTL (closes #2923) (#2953) (@houko) - Named shared workspaces + identity file isolation (#2958) (@houko) - Add notify_owner tool + owner_notice output boundary (#2965) (@houko) - Moonshot/Kimi file upload support via /v1/files (#2966) (@houko) - Download channel files to disk for agent access (#2972) (@houko) - Session_key dispatch log + boot self-test for channel scoping (#2973) (@houko) ### Fixed - Drop ellipsis-terminated preambles without tool_use as silent (#2617) (@f-liva) - Suppress NO_REPLY sentinel in streaming bridge, cron, and auto-reply (#2743) (@DaBlitzStein) - Make split_message HTML-tag-aware for Telegram (#2760) (@DaBlitzStein) - Auto-inject sender peer_id into cron jobs + delegation trust prompt (#2869) (@DaBlitzStein) - Route trigger-fired responses to agent's home channel (closes #2872) (#2952) (@houko) - Render real chat message timestamps on resume (closes #2934) (#2954) (@houko) - Apply assignee_match:self filter to task_posted triggers (closes #2924) (#2955) (@houko) - Inject bot identity into reply_precheck classifier (#2960) (@houko) - Sanitize bot_name in classify_reply_intent prompt; add unit tests (#2961) (@houko) - Tolerate tool_call_id collisions across turns in session_repair (#2962) (@houko) - Inject RELAY prompt only on explicit owner intent (#2967) (@houko) - Add missing timestamp field in session_repair Message structs (#2968) (@houko) - Fix all missing timestamp fields and incomplete test stubs (#2969) (@houko) - Read peer_id from job_json in cron_create (#2970) (@houko) - Recover Signal session when upsert delivers null payload (#2971) (@houko)

LibreFang 2026.4.22

2026-04-22T00:00:00Z

_No notable changes._

LibreFang 2026.4.21

2026-04-21T00:00:00Z

### Added - Complete trigger feature — persistence, CRUD API, CLI subcommands, dashboard UI (#2827) (#2830) (@houko) - Add account_id to channel_send for explicit multi-bot routing (#2845) (@houko) - Add per-agent auto_evolve flag to skip background skill review (#2846) (@houko) - Implement MCP Roots capability (#2847) (@houko) ### Fixed - Correct query invalidation and missing data flow across mutations (#2770) (@leszek3737) - Harden workflow save and draft state (#2781) (@leszek3737) - Align mutation flows across config channels goals and hands (#2782) (@leszek3737) - Unify dashboard query hooks and flow guards (#2783) (@leszek3737) - Exempt Unix/Slack-style timestamps from PII phone check (#2795) (@neo-wanderer) - Change wizard default ollama model to gemma3:4b (#2811) (@houko) - Strip empty assistant messages unconditionally (#2812) (@houko) - Auto-delete At-schedule jobs after execution (#2808) (#2814) (@houko) - Reimplement apply_seccomp_allowlist with libc::SYS_* constants (#2817) (@houko) - Allow dashboard static assets through auth gate (#2824) (@leszek3737) - Force wildcard bind for api_listen in Docker (#2825) (@leszek3737) - Resolve channel_bridge test deadlock that blocked CI for 6h (#2829) (@houko) - ChatPage — type safety, cache correctness, cleanup (#2832) (@leszek3737) - Correct event sequence in show_progress=false test (#2834) (@houko) - Exempt dashboard and static paths from GCRA rate limiter (#2835) (@houko) - Use main as default branch for ~/.librefang git repo (#2837) (@houko) - Task_claim() now matches assigned_to by name as well as UUID (#2844) (@houko) - Dashboard refresh no longer drops history — unify webui session with canonical (#2848) (@houko) - Type-safety and RC-safe fixes (#2849) (@leszek3737) - Unbreak --all-features build + stop warning on local LLM providers (#2850) (@houko) - Per-job session_mode override to fix context accumulation (#2647) (#2851) (@houko) - Proactive extraction loses JSON mode through fork path + log noise cleanup (#2852) (@houko) ### Changed - RC cleanup for ModelsPage (#2833) (@leszek3737) - Relocate config backups under ~/.librefang/backups/ (#2838) (@houko) - Move stray state/log files out of ~/.librefang root (#2840) (@houko) ### Documentation - Add unofficial wiki link and DeepWiki badge to READMEs (#2821) (@leszek3737) ### Maintenance - Run Windows and macOS tests on affected crates for every Rust PR (#2819) (@houko) - Follow-up cleanup from #2783 review (#2820) (@houko) - Ignore rust_out build artifact (#2836) (@houko)

LibreFang 2026.4.20

2026-04-20T00:00:00Z

### Added - Canonical silent-response primitive, end the NO_REPLY literal leak (#2470) (@f-liva) - Gate /dashboard/* behind auth + tailwind v4 renames (#2785) (@houko) - Add stop button to interrupt in-flight agent streams (#2787) (@neo-wanderer) - Add native Cohere driver (#2791) (@houko) - Show tool execution progress in channel replies (#2792) (@houko) - Finish channel-progress — universal coverage, Telegram fix, show_progress, i18n, prettify, dashboard parity (#2793) (@houko) - Redesign `librefang status` for layered visibility (#2799) (@houko) - Unify create/edit modals + inline rename (#2800) (@houko) ### Fixed - Make extract_categories config drive LLM prompt categories (#2761) (@neo-wanderer) - Sync terminal health and active window state (#2777) (@leszek3737) - Clear history consistently and refresh model state (#2780) (@leszek3737) - Align shared query flows for MCP, skills, and workflows (#2784) (@leszek3737) - Route comms_task through kernel wrapper; surface task system events (#2789) (@neo-wanderer) - Rewrite /install to /install.sh for CLI clients (#2794) (@houko) - Stop writing PATH into the wrong rc file (#2796) (@houko) - Auto-activate PATH after installation (#2797) (@houko) - Bypass auth for loopback connections (#2802) (@houko) - Drop stray from #2800 modal refactor (#2803) (@houko) - Surface reload error to dashboard instead of opaque 'saved but reload failed' (#2805) (@houko) - Validate config BEFORE writing TOML so failed saves don't corrupt the file (#2806) (@houko) ### Documentation - Clarify session_mode scope — cron/channels/forks ignore it (#2790) (@neo-wanderer) ### Maintenance - Split PR/main pipelines; compute affected crates precisely (#2801) (@houko) - Merge release-* workflows into one (keep notify) (#2804) (@houko)

LibreFang 2026.4.19

2026-04-19T00:00:00Z

### Added - Add auto-dream per-agent background memory consolidation (#2750) (@houko) - Trigger on AgentLoopEnd hook, scheduler becomes backstop (#2755) (@houko) - Derivative LLM calls reuse parent's prompt cache (#2767) (@houko) ### Fixed - Show Provider before Model in Config default_model section (#2749) (@houko) - Add peer_id to cron jobs for peer-scoped memory access (#2759) (@DaBlitzStein) - Match ImageFile in vision dispatch gates (#2762) (@DaBlitzStein) - Default api_listen to 127.0.0.1:4545 for local-only startup (closes #2766) (#2769) (@houko) - Clear stale TOTP banners, refetch status on reset, localize error messages (#2771) (@leszek3737) - Fix 12 UI bugs across scheduler, sessions, memory, models, plugins, providers, runtime, workflows (#2772) (@leszek3737) - Gate Duration import with cfg(unix) for Windows CI (#2773) (@houko) - Harden canvas workflow recovery and related UI state (#2774) (@leszek3737) - Derive 'connected' from health state + fix catalog card overflow (closes #2738) (#2775) (@houko) - Align workflow mutation invalidation (#2778) (@leszek3737) ### Documentation - Fix stale documentation references (#2720) (@leszek3737) ### Maintenance - Replace cloudflare/wrangler-action with direct npx wrangler calls (#2740) (@houko) ## [Unreleased] ### Changed - Default `api_listen` flipped from `0.0.0.0:4545` to `127.0.0.1:4545` (loopback-only). New installs are local-only by default; set `api_listen = "0.0.0.0:4545"` to expose on LAN/remote. Affects `librefang init`, the dashboard's init endpoint, and `librefang.toml.example`. `librefang start` with an explicit `--config ` that doesn't exist now prints a clear `librefang init` hint instead of failing obscurely. (#2766) - **iOS minimum supported version raised from 14.0 to 16.0.** Required by the Tauri 2 mobile toolchain that the new mobile CI builds against. Devices on iOS 14 or 15 (iPhone 6s, original iPhone SE, iPad Air 2 and similar) will no longer be able to install the LibreFang mobile app once mobile bundles ship. Affects only the iOS app — the desktop and Android builds are unchanged. (#3970) ### Security - **Channel webhook HMAC verification is now mandatory** for Messenger, LINE, Teams, Viber, and DingTalk. Previously, missing signature headers were silently bypassed; they now return `400 Bad Request`, and signature mismatches return `401 Unauthorized`. **Action required if you operate any of these channels:** - **Messenger** — set `MESSENGER_APP_SECRET` to your Facebook App Secret (the new `app_secret_env` field in `[channels.messenger]` defaults to this). If unset, signatures are skipped with a startup warning and the endpoint stays unauthenticated — production should always set it. - **Teams** — set `TEAMS_SECURITY_TOKEN` to the base64 outgoing-webhook security token from the Teams portal (the new `security_token_env` field in `[channels.teams]`). Same fallback semantics as Messenger. - **LINE / Viber / DingTalk** — no new env vars, but probes that don't carry the platform's signature header (curl, monitoring health checks pointed at the webhook path) will now return 4xx instead of 200. - **Outbound `[channels.webhook] callback_url` is SSRF-guarded.** Adapters refuse to start if the URL resolves to a private (`10/8`, `172.16/12`, `192.168/16`), CGN (`100.64/10`), loopback (`127/8`, `::1`), link-local, multicast, or cloud-metadata range. Catches IPv6 short forms like `[::]`, IPv4-mapped (`[::ffff:127.0.0.1]`), NAT64, and trailing-dot FQDNs. **Action required**: local dev setups using `callback_url = "http://127.0.0.1/..."` must switch to a public tunnel (ngrok, cloudflared) or omit `callback_url`. (#3942)

LibreFang 2026.4.18

2026-04-18T00:00:00Z

### Added - Forked agent pattern: kernel exposes `run_forked_agent_streaming(agent_id, prompt, allowed_tools)` for derivative LLM calls that share the parent turn's system + tools + message prefix (Anthropic prompt cache alignment) without persisting the derivative's messages into the canonical session. Anthropic driver's `cache_control` extended from system-only to cover both the last tool block (system + tools prefix) AND the last content block of the last message (full conversation prefix), giving forks near-full cache coverage. Dashboard settings page now surfaces cache-hit rate and per-dream cost so the forkedAgent savings are visible. Proactive-memory `LlmMemoryExtractor` migrated to the forkedAgent pattern: a new trait method `extract_memories_with_agent_id` routes the extraction LLM call through `KernelHandle::run_forked_agent_oneshot` (a new trait method that drives a single-turn fork and returns the final text), sharing the parent agent's `(system + tools + messages)` cache key. The extraction-specific system prompt is embedded into the fork's user message rather than replacing the agent's system prompt, so cache alignment holds. Fall back to a standalone `driver.complete()` with `prompt_caching = true` when no kernel handle is installed (tests / rule-based extractor / fork failure) so system-prompt caching still applies. Kernel wires the extractor's weak handle inside `set_self_handle` — first call only, matching the auto-dream hook idempotency pattern. Migrates auto-dream off its previous `SenderContext { channel: "auto_dream" }` side-channel pattern — dreams now fork from the canonical session and the kernel-side `channel == AUTO_DREAM_CHANNEL` tool filter is replaced by runtime `LoopOptions::allowed_tools` enforcement at tool execute time (request schema stays byte-identical to parent for cache alignment, model's `tool_use` for disallowed tools returns synthetic error). Agent loop adds `LoopOptions { is_fork, allowed_tools }` threaded through; fork turns skip `save_session_async` and add `"is_fork": true` to `AgentLoopEnd` hook context data so subscribers can filter fork events. Auto-dream's own hook filters fork turns to avoid dream-triggers-dream recursion. (@houko) - Auto-dream: per-agent background memory consolidation with four-layer gating (global / per-agent opt-in / time / session count / file lock). Triggered event-driven from the `AgentLoopEnd` hook (fires the moment an agent finishes a turn) with a sparse daily backstop scheduler for opted-in agents that never turn. Includes web dashboard toggle card, TUI Dashboard strip, `[auto_dream]` config section, `DreamConsolidation` audit events with token and cost capture, runtime tool allowlist enforcement, and `GET/POST/PUT /api/auto-dream/status|trigger|abort|enabled` endpoints. (#2750) (@houko) ### Maintenance - Drop bogus npm cache config on setup-node (#2736) (@houko)

LibreFang 2026.4.15

2026-04-15T00:00:00Z

### Added - Add LIBREFANG_DASHBOARD_EMBEDDED_ONLY env var to pin dashboard to embedded assets (#2520) (@neo-wanderer) - Add TOTP scope selector in Settings (#2526) (@houko) - Add section tab switcher to config category pages (#2532) (@houko) - Add voice input button to ChatPage (#2533) (@houko) - Swap tab bar and page header positions in config pages (#2534) (@houko) - Polish config page layout and UX (#2535) (@houko) - Step-by-step provider creation wizard (#2544) (@houko) ### Fixed - Scope telegram sessions per chat_id to prevent context leakage (#2349) (#2522) (@DaBlitzStein) - Honour silent flag in KernelBridgeAdapter sender methods (#2521) (#2523) (@DaBlitzStein) - Use is_some_and instead of map_or in webchat asset_path check (#2525) (@houko) - Move TOTP scope to ConfigPage via schema (#2527) (@houko) - Restore ready-for-review when blockers are cleared (#2528) (@houko) - Fall back to npm when pnpm is unavailable in dev command (#2529) (@houko) - Check review state before clearing needs-changes on push (#2530) (@houko) - Remove needless borrow in serde_json::to_value call (#2531) (@houko) - Show disabled mic button when STT not configured (#2536) (@houko) - Fix stale state bugs in provider config modal (#2537) (@houko) - Move field description to label column (#2538) (@houko) - Show field description below input/toggle (#2539) (@houko) - Save API key on provider creation and show remove button for all providers (#2540) (@houko) - Improve provider auto-detection accuracy and UX (#2542) (@houko) - Remove orphaned doc comment causing clippy failure on main (#2543) (@houko)

LibreFang 2026.4.14

2026-04-14T00:00:00Z

### Added - Pass image blocks to CLI via @path references (#2331) (@f-liva) - MCP OAuth discovery for Streamable HTTP transport (#2346) (@neo-wanderer) - Add require_auth_for_reads to lock down dashboard reads (#2398) (@houko) - Per-call deep-thinking toggle and reasoning display (#2423) (@houko) - Add audit.anchor_path to redirect the tip-anchor file (#2442) (@houko) - Enrich registry cards with manifest metadata (#2452) (@houko) - Channel scoping enforcement, proactive LID, heartbeat watchdog, jittered backoff (#2462) (@f-liva) - PR review state and issue response tracking labels (#2471) (@houko) - Multi-page configuration editor under Configuration nav group (#2473) (@houko) - Group addressee detection — stop responding when not actually spoken to (#2480) (@f-liva) - Per-provider cost/token limits (#2316) (#2482) (@houko) - Add qwen3.6-plus from coding plan (#2494) (@joshuachong) - Add echo tracker to drop our own messages reflected back (#2498) (@f-liva) ### Fixed - Transcode .oga to .ogg before Whisper transcription (#2386) (@f-liva) - Relax brittle alibaba-coding-plan model count assertion (#2388) (@houko) - Block SSRF via IPv4-mapped IPv6 addresses (#2396) (@houko) - Reject path traversal in agent template name param (#2397) (@houko) - Require trusted_manifest_signers for signed manifests (#2407) (@houko) - Make NonceTracker check_and_record atomic and bounded (#2408) (@houko) - Block SSRF via NAT64 well-known prefix (64:ff9b::/96) (#2409) (@houko) - Stop leaking sandbox watchdog threads (#2410) (@houko) - Extend IPv4-mapped IPv6 SSRF guard to remaining call sites (#2411) (@houko) - Clippy regressions from refactor splits (#2404, #2406) (#2412) (@houko) - GCRA rate limiter never honoured per-key token exhaustion (#2413) (@houko) - Strip parent env before host_shell_exec spawns child (#2417) (@houko) - Tighten upload MIME allowlist to match SECURITY.md (#2419) (@houko) - Split_message panic on multi-byte UTF-8 at boundary (#2285) (#2420) (@houko) - Add default connect/read timeouts to shared HTTP client (#2340) (#2421) (@houko) - Lock Owner-only writes away from Admin-role API keys (#2422) (@houko) - Copy button silently failing in non-secure contexts (#2424) (@houko) - At schedules in the past no longer fire forever (#2337) (#2425) (@houko) - Task_claim accepts agent name in addition to UUID (#2330) (#2427) (@houko) - Emit stub tool_results when batch is interrupted (#2381) (#2428) (@houko) - Actually extract WWW-Authenticate from rmcp AuthRequired (#2429) (@houko) - Hot-reload of agent.toml updates ResourceQuota immediately (#2317) (#2430) (@houko) - Add external tip anchor to audit log to detect full rewrites (#2431) (@houko) - Default delivery to LastChannel instead of None (#2338) (#2432) (@houko) - Session_repair phase 3 preserves tool-call boundaries (#2353) (#2433) (@houko) - Claude_code fails fast when agent has tools (#2314) (#2434) (@houko) - Wire audit log through with_db_anchored by default (#2436) (@houko) - Use full viewport width for page content (#2439) (@houko) - Enforce capability inheritance at spawn_agent_inner (#2440) (@houko) - Terminal WebSocket rejected local-dev daemons with no api_key (#2441) (@houko) - Break Feishu bot self-echo loop (#2435) (#2443) (@houko) - Extend taint-sink checks to agent_send and web_fetch body/headers (#2444) (@houko) - Terminal WebSocket froze after ~10 keystrokes from per-message cap (#2445) (@houko) - Cap chat message bubble width for readability (#2446) (@houko) - Taint-scan MCP tool-call arguments before send (#2447) (@houko) - Derive require_auth_for_reads from api_key when unset (#2448) (@houko) - Make overview stats cards responsive at md breakpoint (#2449) (@houko) - Tighten recent agents grid and widen running hand chips (#2450) (@houko) - Repair mobile layout breakage across pages (#2451) (@houko) - Tighten card grid breakpoints across pages (#2453) (@houko) - Revert issue auto-label body scan, keep keyword expansion (#2457) (@houko) - Match camelCase/snake_case keywords in issue auto-label (#2461) (@houko) - Scope canonical context injection per session to stop cross-chat leak (#2464) (@f-liva) - Stop killing unrelated process groups in tree-kill path (#2472) (@houko) - Bridge LibreFang tools to claude_code driver via MCP config (#2314) (#2478) (@houko) - Scope canonical context injection per session to stop cross-chat leak (#2464) (#2490) (@houko) - Wire MCP bridge end-to-end for claude_code (#2314) (#2495) (@houko) - Use direct libc::kill syscall to prevent Ubuntu CI SIGTERM (#2497) (@houko) ### Changed - Extract http_client into librefang-http shared crate (#2389) (@houko) - Extract metering into librefang-kernel-metering subcrate (#2395) (@houko) - Extract oauth flows into librefang-runtime-oauth subcrate (#2400) (@houko) - Extract mcp into librefang-runtime-mcp subcrate (#2403) (@houko) - Extract drivers and llm_driver trait into subcrates (#2404) (@houko) - Extract wasm sandbox and kernel-handle trait into subcrates (#2405) (@houko) - Extract hand/template router into librefang-kernel-router subcrate (#2406) (@houko) - Remove bare SignedManifest::verify() and inline it as private (#2437) (@houko) - Rename librefang-runtime-drivers to librefang-llm-drivers (#2467) (@houko) - Extract pure helpers and tests out of kernel.rs (#2469) (@houko) ### Documentation - Describe prompt-injection scanner as a heuristic (#2399) (@houko) - Audit chain is tamper-evident only against partial edits (#2415) (@houko) - Narrow the secret-zeroization claim to its actual scope (#2416) (@houko) - Describe taint tracking as a two-sink pattern match (#2426) (@houko) - Document additive penalty assumption in fallback recover (#2465) (@f-liva) ### Maintenance - Stabilize load_endpoint_latency against shared-runner jitter (#2418) (@houko) - Remove stray empty .codex marker file (#2454) (@houko) - Broaden issue auto-label coverage and add backfill (#2455) (@houko) - Refresh dashboard screenshot and drop unused images (#2456) (@houko) - Address houko follow-ups on oga transcode (#2459) (@f-liva) - Tidy repo metadata and remove stale api-docs (#2466) (@houko) - PR conflict/CI-failure detection and issue status labels (#2481) (@houko) - Sync Cargo.lock with librefang-api toml_edit dep (#2500) (@houko) - Sync Cargo.lock after librefang-llm-driver dep addition (#2501) (@houko)

LibreFang 2026.4.13

2026-04-13T00:00:00Z

### Added - Allow editing hand agent model settings from agents page (#2335) (@leszek3737) - Add config-driven session_mode for agent triggers (#2341) (@neo-wanderer) - Telegram rich media, polls, interactive commands, and channel_send tool (#2356) (@leszek3737) ### Fixed - Decryption retry, streaming tag leak, session isolation (#2217) (@f-liva) - Inherit kernel default_model instead of hardcoded Anthropic (#2299) (@houko) - Per-agent loading state so streaming one agent doesn't block others (#2324) (@houko) - Write MCP server config as TOML table, not stringified JSON (#2327) (@houko) - Load secrets.env autonomously at boot time (#2359) (@f-liva) - Prevent zombie processes on shutdown (#2360) (@f-liva) - Refuse direct DELETE on hand-spawned agents + clarify revert warning (#2361) (@houko) - Normalize MIME type parameters before allowlist check (#2362) (@f-liva) - Resolve LID JIDs to phone numbers for owner detection (#2363) (@f-liva) - Harden poll_options parsing and poll context cleanup (#2364) (@houko) - Deterministic prompt context ordering and raise truncation cap (#2365) (@houko) - Stop Qwen driver from leaking raw JSON into chat (#2366) (@f-liva) - Let FallbackDriver recover from transient unhealthiness (#2367) (@f-liva) - Clear stale per-agent overrides on provider switch (#2371) (@neo-wanderer) - Scrub NO_REPLY sentinel in every reply path (#2373) (@f-liva) - Restore /message/send-audio endpoint accidentally removed in #2217 (#2376) (@f-liva) - Support "date" metric format and drop ureq from cli (#2382) (@houko) ### Performance - Shrink dev debug info to line-tables-only (#2378) (@houko) ### Maintenance - Split Docker image and deploy status (#2323) (@houko) - Fix max_tokens assertions after pure-text short-circuit (#2325) (@houko) - Strengthen telegram sanitizer coverage (#2334) (@leszek3737) - Fix rustfmt on upsert_mcp_server test assert (#2358) (@houko) - Replace cat with sleep in process_manager tests to fix flake (#2375) (@houko) - Skip security and install-smoke on unrelated PRs (#2377) (@houko) - Apply cargo fmt to runtime drivers (#2380) (@houko) ## [Unreleased] ### Added - Config-driven session mode for agent triggers (`session_mode = "new" | "persistent"`) — per-agent default with per-trigger override ### Security - **BREAKING**: `require_auth_for_reads` now defaults to *enabled* whenever any form of authentication is configured (`api_key`, `user_api_keys`, or dashboard credentials). Previously the flag had to be set explicitly, leaving read endpoints open even on instances with an `api_key`. Operators who deliberately want open reads on an authenticated instance (e.g. behind a trusted reverse proxy) must now set `require_auth_for_reads = false` in `config.toml`. A boot-time INFO log records when the flag is auto-enabled. (#2448)

LibreFang 2026.4.11

2026-04-11T00:00:00Z

### Added - Add WebSocket terminal with PTY backend and xterm frontend (Phase 1) (#2229) (@leszek3737) - Claude Code CLI profile rotation for rate-limit resilience (#2249) (@f-liva) - Add MCP Servers management page (#2278) (@houko) - Raise MSRV to 1.94.1 and keep stable toolchain (#2302) (@houko) - Uninstall hand (#2312) (@houko) ### Fixed - Change Docker setup to fix permissions for LIBREFANG_HOME (#2240) (@Cruel) - Also ignore secrets.env (dashboard-managed env file) (#2248) (@DaBlitzStein) - Localize agent template copy for zh users (#2257) (@houko) - Restore approval context and dashboard auth flows (#2272) (@houko) - Exclude Hand sub-agents from channel routing fallback (#2276) (@houko) - Accept claude-code (hyphen) in CLI profile rotation guard (#2284) (@f-liva) - Replace --verbose with --include-partial-messages for qwen driver (#2290) (@f-liva) - Add missing cli_profile_dirs to DefaultModelConfig literals (#2296) (@houko) - Delegate first-boot config to librefang init (#2297) (@houko) - Scan workspaces/ dir to persist locally-installed hands across boot (#2298) (@houko) - Hide delete button for built-in providers, flag custom (#2300) (@houko) - Mark manifest mut in parse_manifest (#2306) (@houko) - Stop middleware path normalization from swallowing GET / (#2307) (@houko) - Preserve pending Telegram updates across daemon restart (#2309) (@houko) - Stop agent loop on pure-text max_tokens overflow (#2310) (@houko) - Make Hands Settings tab actually editable (#2311) (@houko) - Wire ConPTY resize on Windows (#2313) (@houko) ### Changed - Harden and optimize Telegram adapter (#2223) (@leszek3737) ### Maintenance - Cover full-path context hook launchers (#2255) (@houko) - Cover wechat and wecom multi-account config parsing (#2258) (@houko) ### Other - Feat(ws) harden terminal websocket follow-ups after #2229 (#2304) (@houko)

LibreFang 2026.4.10

2026-04-10T00:00:00Z

### Added - Per-channel session isolation via deterministic UUID v5 (#2097) (@f-liva) - Save channel images as files instead of inline base64 (#2098) (@f-liva) - TOTP second-factor for critical tool approvals (#2131) (@houko) - Proper resource composition for hand agents (#2133) (@houko) - Add extra_params support for openai compatible model (#2181) (@houko) - Add config export/backup endpoint and UI button (#2186) (@houko) - Prefill TOML editor from template selection (#2187) (@houko) - Add per-channel auto-routing with configurable strategies (#2189) (@houko) - Allow hooks to access vault secrets via allowed_secrets (#2216) (@houko) - Add [config] section support to plugin.toml (#2218) (@houko) - Add [[requires]] system binary checks to plugin.toml (#2219) (@houko) ### Fixed - Detect "[no reply needed]" as silent response (#2093) (@f-liva) - Harden agent loop tool flow and trim handling (#2135) (@leszek3737) - Timezone-aware schedule creation (#2138) (@f-liva) - Replace librefang.dev with librefang.ai (#2147) (@houko) - Glob-match declared tools and auto-promote shell_exec exec_policy (#2148) (@houko) - Persist mcp server updates in patch agent (#2151) (@TechWizard9999) - Use codex exec for codex cli driver (#2153) (@TechWizard9999) - Improve Claude Code detection for keychain auth and non-login shells (#2166) (@x86txt) - Show active agent count instead of total in overview card (#2170) (@DaBlitzStein) - Handle SkillHub search response format with proper headers (#2171) (@DaBlitzStein) - Suppress CMD window flash on Windows (#2159) (#2176) (@houko) - Resolve hand.toml agent scan conflict (#2136) (#2177) (@houko) - Parameter errors trigger self-correction not user report (#2144) (#2178) (@houko) - Resolve pre-existing clippy and test compile failures (#2180) (@houko) - Multi-bot Telegram routing uses account_id, not first-match on allowed_users (#2183) (@houko) - Resolve build errors and clippy warnings (#2184) (@houko) - Skip auto-init when piped via curl, prompt user to run manually (#2190) (@houko) - Clean up post-install messaging for piped installs (#2192) (@houko) - Replace as_deref() with as_ref() for ChannelOverrides in bridge.rs (#2193) (@houko) - Add missing extra_body field to make_completion_request (#2197) (@houko) - Remove dead completion_timeout_override and build_completion_request (#2198) (@houko) - Derive Default for PluginManifest (#2205) (@houko) - Add INFO logs for all ingest hook success paths (#2213) (@houko) - Reduce agent count display lag on state changes (#2215) (@houko) - Decryption retry, streaming tag leak, session isolation (#2217) (@f-liva) - Filter tool_use/tool_result blocks from chat rendering (#2220) (@f-liva) - Resolve default provider in agent detail endpoint (#2221) (@DaBlitzStein) - Resolve default provider before creating driver (#2222) (@DaBlitzStein) - Add error handling to channel config dialog (#2224) (@DaBlitzStein) - Default to unconfigured tab when no channels are set up (#2225) (@DaBlitzStein) - Propagate ClawHub/Skillhub errors instead of returning 200 OK with empty items (#2231) (@DaBlitzStein) - Fix compile errors and rustfmt from Custom variant merge (#2234) (@houko) - Show embedding status ok when fts_only mode is active (#2236) (@houko) - Rustfmt formatting in snapshot handler (#2237) (@houko) - Rustfmt formatting in config routes (#2238) (@houko) - Merge extra_body into JSON Value to avoid duplicate keys (#2239) (@shilkazx) - Scope RwLockReadGuard before await in dashboard_snapshot (#2241) (@houko) - Increase dark theme surface opacity for readable dropdowns (#2242) (@houko) - Always load marketplace skills even without search keyword (#2243) (@houko) ### Changed - Typed enums, O(1) indexes, and typed persistence v4 (#2161) (@leszek3737) ### Maintenance - Apply rustfmt formatting across bridge, router, kernel, system (#2195) (@houko) - Remove extra blank line in agent_loop.rs (#2203) (@houko) - Remove mempalace-indexer from contrib — moved to registry (#2247) (@houko)

LibreFang 2026.4.7

2026-04-07T00:00:00Z

### Fixed - Resume agent loops after approval without blocking (#2101) (@leszek3737) - Skip Discord notification when release workflows are cancelled (#2129) (@houko) - Embed dashboard in release binaries (#2132) (@houko) ### Maintenance - Add desktop build/dev recipes to justfile (#2134) (@houko)

LibreFang 2026.4.6

2026-04-06T00:00:00Z

### Added - Hot-reload skills dir and per-agent manifest (#2069) (@houko) - Unify full-section empty/error states (#2088) (@houko) - Focus trap + aria-modal + more n-shortcut coverage (#2092) (@houko) - Add send-audio endpoint for voice notes and audio files (#2099) (@f-liva) - Language-agnostic hook runtime (V / Go / Deno / Node / native) (#2100) (@houko) ### Fixed - Allow tool retry on failure instead of early loop termination (#2065) (@neo-wanderer) - Sync openclaw/openfang with current KernelConfig schema (#2066) (@houko) - Stop stale messages_before index from breaking auto_memorize & append_canonical (#2068) (@houko) - Agent_send/kill fall through to name lookup for stale UUIDs (#2070) (@houko) - Reject missing required tool params instead of silent empty (#2071) (@houko) - Surface silent session-cleanup failures and panic on empty chunks (#2072) (@houko) - Return 404 for missing agents and reject malformed target_agent_id (#2073) (@houko) - Log when webhook/dingtalk bridge drops incoming messages (#2074) (@houko) - Surface agent tick panics instead of silent join drop (#2075) (@houko) - Emit skills/workspace/tool_blocklist during OpenClaw import (#2076) (@houko) - Providers.rs persistence failures + expect() panic (#2077) (@houko) - Surface silent DB errors and wrap merge updates in tx (#2078) (@houko) - Surface episodic memory persist failures in agent_loop (#2079) (@houko) - Sanitize user-controlled identity fields in prompt builder (#2080) (@houko) - Reload path must clamp bounds and clamp max_cron_jobs=0 (#2081) (@houko) - Close SSRF via redirect + URL-encoding bypass in taint (#2082) (@houko) - Route media tools through workspace sandbox (#2083) (@houko) - Guard sandbox ptr arithmetic with checked_add (#2084) (@houko) - ChatPage session-cache save effect + tool call keys (#2085) (@houko) - Cascade agent-scoped tables on remove_agent (#2086) (@houko) - Authorize cron_cancel + cap knowledge_query depth (#2087) (@houko) - Use PAT for release creation so dashboard-build fires (#2094) (@houko) - Suppress error messages in groups, show rate-limit in DMs only (#2095) (@f-liva) - Auto-close unclosed HTML tags, plain-text fallback, and reply-to photo support (#2096) (@f-liva) - Drop Ubuntu RUST_TEST_THREADS to 1 (#2117) (@houko) - Unify agent manifest path on workspaces/agents/ (#2118) (@houko) ### Changed - Align URL hierarchy with sidebar nav groups (#2119) (@houko) ### Maintenance - Fix test_image_analyze_missing_file after sandbox wiring (#2103) (@houko) - Ignore plugin scaffold templates (#2120) (@houko) ### Reverted - V2026.4.6 stable release (was meant to be beta15) (#2126) (@houko)

LibreFang 2026.4.5

2026-04-05T00:00:00Z

### Added - Add inline tool use display to chat UI (#2031) (@neo-wanderer) - Support username and @username in allowed_users filter (#2036) (@leszek3737) - Add alibaba coding plan as provider (#2040) (@joshuachong) - Add hidden models — hide/unhide models from selectors (#2045) (@leszek3737) - HITL notification engine, batch ops, modify-and-retry, audit log (#2046) (@houko) - Add media generation page (#2051) (@houko) - Redesign Hands page with running strip and richer cards (#2052) (@houko) - Redesign Hands detail modal with hero, action bar, metrics strip (#2053) (@houko) - Polish Hands list — grid skeleton, empty states, degraded (#2054) (@houko) - Per-channel command policy for public-facing bots (#2063) (@houko) ### Fixed - Stop embedding dashboard artifacts in release commits (#2039) (@houko) - Remove tracked static/react/ build artifacts from git (#2041) (@houko) - Trigger dashboard build on release publish (#2043) (@houko) - Strip provider prefix from agent fallback_models (#2047) (@houko) - Ensure static/react dir exists for include_dir! (#2048) (@houko) - Defer WebSocket close until connection is established (#2050) (@houko) - Hands detail modal tab bar height, underline, and schedules label (#2055) (@houko) - Remove count pills from Hands detail tabs to guarantee equal height (#2056) (@houko) - Auto-wire self handle in streaming path for inter-agent tools (#2061) (@houko) - Scope per-turn recall by peer_id to stop cross-user leaks (#2062) (@houko) ### Documentation - Update dashboard build references after static/react removal (#2042) (@houko) - Clarify routing lives in agent manifest, not config.toml (#2060) (@houko) ### Maintenance - Fix 20 pre-existing TypeScript errors (#2049) (@houko)

LibreFang 2026.4.4

2026-04-04T00:00:00Z

### Added - Interactive model switcher dropdown in connection bar (#1995) (@neo-wanderer) - Custom model management, workflow scheduling, and HandsPage fixes (#2028) (@houko) - Wire up channel test/reload and session labels (#2030) (@houko) - Serve dashboard from runtime directory with auto-sync (#2032) (@houko) ### Fixed - Prevent duplicate TOML keys during config upgrade (#2025) (@houko) - Unify scheduling system, improve dashboard and hand UX (#2026) (@houko) - Sync Cargo.lock for flate2/tar dependencies (#2034) (@houko)

LibreFang 2026.4.3

2026-04-03T00:00:00Z

### Fixed - Use plain reqwest client in integration tests (#2000) (@houko) - Add elevenlabs support to API key test endpoints (#2005) (@Chukwuebuka-2003) - Add retry logic to release asset upload steps (#2007) (@houko)

LibreFang 2026.4.2

2026-04-02T00:00:00Z

### Added - Press 'r' in just dev to git pull and rebuild (#1949) (@houko) - Inline session switcher in chat (#1953) (@houko) - Dev hotkeys and auto-pull (#1955) (@houko) ### Fixed - Expose cleanup_orphan_sessions on MemorySubstrate (#1943) (@houko) - Skip non-GET requests in service worker cache (#1944) (@houko) - Route hand agent workspace to hands/ instead of agents/ (#1945) (@houko) - Preserve depends_on when instantiating templates (#1946) (@houko) - Add proxy timeout and WebSocket support for dev server (#1947) (@houko) - Respect usage_footer config in chat message footer (#1948) (@houko) - Git pull from origin/main in dev hotkey (#1950) (@houko) - Validate provider keys and model availability on boot (#1951) (@houko) - Use fetch+rebase for dev 'r' hotkey (#1952) (@houko) - Remove unused binary_clone variable (#1954) (@houko) - Match usage_footer values to backend snake_case (#1956) (@houko) - Serialize usage_footer with serde instead of Debug format (#1957) (@houko) - Point skillhub API to skillhub.tencent.com (#1958) (@houko) - Skillhub install via COS direct download (#1959) (@houko) - Remove hardcoded default models and add model availability probe (#1960) (@houko) - Install FangHub skills from local registry instead of GitHub (#1961) (@houko) - Infer provider from model name in fallback resolution (#1962) (@houko) - FangHub install and search use local registry (#1963) (@houko) - Mark unreachable local providers as unavailable (#1964) (@houko) - Assistant agent model not updated when config changes (#1965) (@houko) - Test provider should check CLI availability before requiring API key (#1966) (@houko) - Local provider status driven by probe, not detect_auth (#1967) (@houko) - Filter hand agents from analytics and telemetry (#1968) (@houko) - Rename plugin source to plugin marketplace in Chinese locale (#1969) (@houko) - Remove install button from plugins page header (#1970) (@houko) - Startup health check respects explicit api_key_env config (#1973) (@houko) ### Changed - Remove bundled system and add per-hand skill install (#1942) (@houko)

LibreFang 2026.4.1

2026-04-01T00:00:00Z

### Added - Add ssrf_allowed_hosts allowlist for web_fetch (#1899) (@houko) - Add embedding provider auto-detection (#1901) (@houko) - Translate built-in agent names in dashboard (#1913) (@houko) ### Fixed - Sync streaming fixes (#1897) (@houko) - Sync config defaults (#1898) (@houko) - Trigger ReloadSkills on skills config TOML changes (#1900) (@houko) - Prevent users=[] conflict with [[users]] array-of-tables (#1904) (@houko) - Fix file_write failed bug when create directory with non-exists … (#1905) (@shilkazx) - Google_tts size check and is_ssml false-positive test coverage (#1906) (@houko) - Prevent NO_REPLY token from leaking in group chats (#1908) (@f-liva) - Resolve symlinked workspace roots on macOS (#1910) (@houko) ### Maintenance - Fetch full tag history so diff link is populated (#1907) (@houko)

LibreFang 2026.3.31

2026-03-31T00:00:00Z

### Fixed - Replace _redirects with _worker.js for SPA routing (#1824) (@houko) - Add auto-init step to Windows installer (#1825) (@houko) - Auto-init on first run for start/chat commands (#1826) (@houko) - Resolve all open issues (#1827 #1828 #1829 #1830 #1832) (#1834) (@houko) - Add missing message_timeout_secs in test DefaultModelConfig (#1835) (@houko) - Add missing message_timeout_secs in DefaultModelConfig initializers (#1836) (@houko) - Remove needless borrow for clippy (Rust 1.94) (#1838) (@houko) ### Documentation - Fix development guide with just usage and dashboard debugging (#1831) (@houko) - Add Windows exe manual install guide (#1833) (@houko) ### Maintenance - Fix workflow trigger issues and add concurrency controls (#1822) (@houko) - Remove redundant web-lint workflow (#1823) (@houko)

LibreFang 2026.3.30

2026-03-30T00:00:00Z

### Added - Add configurable IMAP email reader (#1322) (@devatsecure) - Add message debounce with shutdown flush (#1684) (@Chukwuebuka-2003) - Convert markdown to WhatsApp formatting (#1733) (@f-liva) - Add WeCom callback mode UI (#1773) (@houko) - Add AGENTS.md for AI assistant context (#1779) (@houko) - Add password change support (#1780) (@houko) - Add registry_mirror for faster marketplace access in China (#1783) (@houko) - Add wildcard pattern support for tool capabilities (#1801) (@houko) - Add voice channel adapter with WebSocket server (#1802) (@houko) - Add DingTalk stream mode support (#1804) (@houko) - Auto-init config and copy example on first just dev (#1808) (@houko) - Add Streamable HTTP transport, custom headers, and browser.enabled config (#1809) (@houko) ### Fixed - Auth bootstrap for protected sessions (#1687) (@TechWizard9999) - Allow Windows absolute paths in secrets.env and config.toml writes (#1770) (@SenZhangAI) - Load full workflow detail after template instantiation (#1772) (@SenZhangAI) - Add event_id dedup to feishu adapter (#1776) (@houko) - Skip disabled agents during background startup (#1777) (@houko) - Stop hiding hand agents from chat sidebar (#1778) (@houko) - Align probe result fields with dashboard (#1781) (@houko) - Handle all HTTP error codes in provider test (#1782) (@houko) - Refresh provider catalog in-place after registry write (#1784) (@houko) - Add versioned migration flow with best-effort fallback (#1785) (@houko) - Improve NO_REPLY detection, raise history limit, preserve user messages (#1787) (@f-liva) - Don't cancel in-progress runs on main branch (#1788) (@houko) - Use per-SHA concurrency group on main to prevent SIGTERM (#1794) (@houko) - Install npm in runtime image (#1799) (@j5bart) - Route Telegram messages to correct agent (#1803) (@houko) - Throttle Ubuntu test to prevent OOM SIGTERM (#1805) (@houko) - Limit nextest to 1 concurrent test binary on Ubuntu (#1807) (@houko) - Respect default_agent in channel message routing (#1810) (@houko) - Propagate group context and @mention detection (#1811) (@houko) - Complete group chat support (P1-P3) (#1812) (@houko) - Use mutable default for non-exhaustive config struct (#1814) (@houko) - Add missing PromptContext fields from WhatsApp group PR (#1816) (@houko) - Re-apply provider URLs after runtime catalog sync (#1818) (@leszek3737) - Remove duplicate is_group/was_mentioned in PromptContext (#1820) (@houko) ### Other - Update dashboard image in markdown (#1746) (@Jengro777)

LibreFang 2026.3.28

2026-03-28T00:00:00Z

### Added - TUI guide for free provider setup on first run (#1731) (@houko) - Add set-as-default button to provider UI (#1753) (@houko) ### Fixed - Use English for shared contacts label (#1732) (@f-liva) - Use live default model for provider auth checks (#1748) (@TechWizard9999) - Hot-reload Wecom channel config without restart (#1754) (@houko) - Use effective default provider instead of hardcoded OpenRouter (#1755) (@houko) - Add parse_mode and sanitization to streaming initial message (#1759) (@f-liva) - Avoid blocking_write panic in daemon on Termux/Android (#1765) (@houko) ### Maintenance - Batch upgrade dependencies (#1752) (@houko)

LibreFang 2026.3.26

2026-03-26T00:00:00Z

### Added - Persist workflow run state to survive daemon restarts (#1657) (@houko) - Add nvidia/nim aliases for nvidia-nim provider (#1660) (@houko) - Sync and serve channel metadata from registry (#1661) (@houko) - Integrate goal system into agent loop and prompt builder (#1663) (@houko) - Migrate MCP stdio transport to rmcp SDK, fix env leak (#1667) (@houko) - Implement all missing hot-reload actions (#1679) (@houko) - Pluggable VectorStore backend with HTTP implementation (#1691) (@houko) - Multimodal memory schema foundation for image indexing (#1692) (@houko) - Add 5 operator-facing config fields (tool_timeout, upload_size, concurrency, call_depth, body_size) (#1709) (@houko) - Add /api/registry/schema endpoint for dashboard form generation (#1715) (@houko) - Add upgrade mode to librefang init (#1723) (@houko) - Replace WeCom app with intelligent bot WebSocket adapter (#1729) (@houko) ### Fixed - Replace unsafe pointer mutation in budget config updates (#1637) (@houko) - Make metering quota check and usage record atomic (#1638) (@houko) - Add TTL-based expiration for A2A task store (#1639) (@houko) - Track background tasks for graceful shutdown (#1640) (@houko) - Use atomic DashMap entry API for agent registry name index (#1641) (@houko) - Replace production panics with error handling (#1642) (@houko) - Support multiple Hand instances with instance-scoped agent IDs (#1643) (@houko) - Auto-patch node-gyp on Termux/Android for better-sqlite3 native build (#1649) (@houko) - Use centralized http_client to avoid rustls-platform-verifier panic on Termux (#1650) (@houko) - Centralize registry sync to prevent parallel git clone races (#1651) (@houko) - Pin DNS resolution to prevent SSRF rebinding attacks (#1653) (@houko) - Add 8 missing fields to strict config validation (#1654) (@houko) - Log warnings for malformed LLM tool call arguments (#1655) (@houko) - Add per-trigger cooldown to prevent event storms (#1656) (@houko) - Resolve WhatsApp gateway config path from $HOME instead of hardcoded /data/ (#1658) (@houko) - Enforce workspace sandbox and tool capability checks (#1665) (@houko) - Dashboard auth dialog never shown when api_key is configured (#1666) (@houko) - Add dropped event monitoring to event bus (#1668) (@houko) - Docker symlink, memory merge, workflow conditions, config test (#1670) (@houko) - Enforce tool call and cost quotas in scheduler (#1671) (@houko) - Apply cache token discount and update model prices (#1672) (@houko) - Implement OAuth refresh token flow (#1673) (@houko) - Replace XOR obfuscation with Argon2 key wrapping (#1674) (@houko) - Make config hot-reload atomic with epoch counter (#1676) (@houko) - Remove dead client field from WebFetchEngine (#1678) (@houko) - Restore backward-compatible agent IDs for single-instance hands (#1680) (@houko) - Re-land SSRF DNS pinning to prevent TOCTOU rebinding attacks (#1681) (@houko) - Budget enforcement, complete API error migration, cache invalidation (#1683) (@houko) - Clippy warnings and rustfmt from recent merges (#1685) (@houko) - Update hand tests for legacy agent ID format (#1686) (@houko) - Sync workflow templates from registry on boot (#1688) (@houko) - Remove workflows from registry sync (kernel handles this separately) (#1689) (@houko) - Webchat responses silently dropped due to stream timeout and missing routing context (#1690) (@houko) - Resolve compilation errors from merged PR conflicts (#1712) (@houko) - Suppress clippy::manual_clamp in clamp_bounds (#1716) (@houko) - Remove dangling doc comment in ws.rs (#1717) (@houko) - Wrap load_templates_from_dir with block_in_place (#1719) (@houko) - Repair test failures from goal system merge (#1720) (@houko) - Recognize all available auth statuses for custom providers in WebUI (#1721) (@houko) - Correct test expectations for metering and workflow collect (#1722) (@houko) - Accept "Failed to resolve" error in Windows capability test (#1725) (@houko) - Auto-detect default LLM provider, fix WeChat QR flashing (#1727) (@houko) ### Changed - Standardize API error response format (#1646) (@houko) - Deduplicate LLM driver request building and fix streaming (#1669) (@houko) - Deduplicate constants and auto-generate user-agent version (#1693) (@houko) - Remove pub const provider URLs, inline in driver registry (#1695) (@houko) - Extract registry cache TTL into configurable RegistryConfig (#1698) (@houko) - Extract API rate limiting constants into RateLimitConfig (#1701) (@houko) - Extract compaction constants into CompactionConfig (#1704) (@houko) - Extract trigger system constants into TriggersConfig (#1705) (@houko) - Extract channel timeout and polling constants into per-channel config (#1707) (@houko) - Move workflow template sync from kernel boot to registry_sync (#1713) (@houko) ### Performance - Cache available_tools computation per agent (#1644) (@houko) ### Maintenance - Extract build_agent_manifest_toml from tool_agent_spawn and test (#1648) (@aimlyo) - Remove bundled integration templates from source tree (#1659) (@houko) - Fix formatting issues caught by CI (#1714) (@houko)

LibreFang 2026.3.25

2026-03-25T00:00:00Z

### Added - TUI multi-select provider menu in deploy script (#1618) (@houko) - Add publish links to SDK release job summary (#1623) (@houko) - Limit-the-degrees-of-freedom-of-agent_spawn (#1624) (@aimlyo) ### Fixed - Read from /dev/tty in deploy script for curl-pipe compatibility (#1616) (@houko) - TUI arrow key navigation crashes due to set -e (#1620) (@houko) - Add -- to grep patterns in release workflows (#1622) (@houko) - Use isolated test dir for model_catalog tests (#1627) (@houko) - Resolve DMG asset name mismatch in Homebrew Cask sync (#1628) (@houko) - Embed contributor avatars as base64 in SVG (#1630) (@houko) - Always tag Docker image as :latest (#1631) (@houko) ### Maintenance - Stop marking beta/rc as GitHub prerelease (#1626) (@houko)

LibreFang 2026.3.24

2026-03-24T00:00:00Z

### Added - Implement depends_on DAG execution for workflow steps (#1440) (@houko) - Add workflow template API endpoints (#1442) (@houko) - Wire thinking model configuration into agent loop (#1443) (@houko) - Mobile responsive + PWA + login + skill output persistence (#1445) (@houko) - Implement session context injection with multiple sources (#1448) (@houko) - Save existing workflow as reusable template (#1449) (@houko) - Add Shell/Bash skill runtime (#1450) (@houko) - Add push messaging API for agents to send to channels (#1451) (@houko) - Add /btw ephemeral side question command (#1452) (@houko) - Add structured output (JSON/JSON Schema) for agents (#1453) (@houko) - Add session export/import for context hibernation (#1454) (@houko) - Configurable heartbeat timeout and pruning per agent (#1455) (@houko) - Cross-session wake via target_agent on triggers (#1456) (@houko) - Add interactive message payloads for Telegram and Slack (#1457) (@houko) - Add PII privacy controls with pseudonymization and redaction (#1458) (@houko) - Tool-level authorization with per-sender and channel-specific policies (#1459) (@houko) - Subagent context inheritance in workflow steps (#1460) (@houko) - Lazy-load LLM driver cache for improved runtime performance (#1461) (@houko) - Add Amazon Bedrock embedding driver with SigV4 signing (#1462) (@houko) - FTS5 full-text session search with API endpoint (#1463) (@houko) - Message injection between tool calls (mid-turn interrupt) (#1464) (@houko) - Render LaTeX in chat (#1467) (@TechWizard9999) - Automatic memory chunking for long documents (#1468) (@houko) - Input sanitizer for prompt injection detection (#1469) (@houko) - Add Android (aarch64) cross-compilation for Termux users (#1470) (@houko) - Time-based memory decay for hierarchical memory management (#1471) (@houko) - File-based input inbox for async external commands (#1472) (@houko) - Interactive approval dialog in dashboard chat and channel events (#1474) (@houko) - Telegram thread-based agent routing (#1475) (@houko) - Pause/resume, busy guard, AgentManifest composition (#1482) (@houko) - Add librefang-testing crate with mock infrastructure (#1483) (@houko) - Show GitHub compare link before version confirmation (#1488) (@houko) - Integrate Skillhub marketplace as second skill source (#1504) (@houko) - Add WeChat personal account adapter via iLink protocol (#1506) (@houko) - Comprehensive build automation CLI with 31 subcommands (#1511) (@houko) - Enhance Hand system with i18n, pause/resume, and dashboard overhaul (#1515) (@houko) - Enable by default, add Grafana, auto-start with Docker (#1520) (@houko) - Multi-agent hand architecture (#1521) (@houko) - Add regex group trigger patterns (#1529) (@TechWizard9999) - Generic media generation drivers (image, TTS, video, music) (#1532) (@houko) - Extend Prometheus metrics and add Grafana dashboards (#1533) (@houko) - Add LTS version support (#1535) (@houko) ### Fixed - Handle paginated /api/agents response (#1233) (@f-liva) - Preserve caption on Telegram voice messages (#1249) (@f-liva) - Detect and retry when LLM skips tool execution for action requests (#1413) (@houko) - Stop agent loop on tool execution failure (#948) (#1415) (@houko) - Complete ChatGPT Responses driver streaming/tool/reasoning mapping (#1405) (#1421) (@houko) - Use 2-digit year in Tauri version for WiX MSI compatibility (#1439) (@houko) - Harden workflow permissions and catalog path validation (#1444) (@SenZhangAI) - Stabilize nodeTypes to fix workflow builder editing (#1447) (@houko) - Harden reconnect and request handling (#1465) (@TechWizard9999) - CI shell injection, clippy warnings, init config, and review findings (#1473) (@houko) - Validate tool_use.input as dict in Anthropic and OpenAI drivers (#1476) (@houko) - Replace plaintext password with Argon2id hashing (#1477) (@houko) - Replace git-based registry sync with HTTP tarball download (#1479) (@houko) - Hand registry race condition, state persistence, and optional requirements (#1481) (@houko) - Resolve clippy errors blocking all PRs (#1486) (@houko) - Consolidate confirmations into single final prompt (#1491) (@houko) - Align chat websocket contract (#1498) (@poruru-code) - Exempt non-autonomous agents from timeout check (#1499) (@houko) - Stamp last_active before LLM call (#1500) (@houko) - Reset last_active on agent restore (#1501) (@houko) - Resolve clippy and compilation errors from merged PRs (#1502) (@houko) - Use tokio::test for callback query tests (#1503) (@houko) - Resolve compilation and clippy errors from recent merges (#1507) (@houko) - Update tool fallback assertions for capability enforcement (#1508) (@houko) - Follow up merged PR regressions (#1514) (@houko) - Use endpoint discovery API for Feishu WebSocket connection (#1518) (@houko) - Gitignore, channel logging, and xtask Windows CI (#1519) (@houko) - Preserve coordinator role and role-bound trigger migration (#1523) (@houko) - Restore --release flag in Dockerfile build (#1524) (@houko) - Eliminate username enumeration timing side-channel (#1525) (@houko) - Replace deterministic session token with random generation (#1526) (@houko) - Prevent path traversal in skill script execution (#1527) (@houko) - Make init_prometheus idempotent for parallel test safety (#1528) (@houko) - Multi-agent parsing compat + registry sync version update (#1530) (@houko) - Gate unix-only test behind #[cfg(unix)] (#1534) (@houko) - Release tool compares against latest tag including prereleases (#1547) (@houko) - Release tool retries commit after formatter hook (#1548) (@houko) - Release tool compares against latest tag including prereleases (#1547) (#1550) (@houko) - Remove unused find_latest_stable_tag in release.rs (#1551) (@houko) ### Changed - Add facade getters and migrate API routes (#1478) (@houko) - Modularize route registration into per-domain routers (#1484) (@houko) - Split monolithic config.rs (5566 LOC) into modular sub-modules (#1485) (@houko) - Registry as catalog, pre-install core content only (#1537) (@houko) - Unified workspaces layout + hand/agent isolation + routing fixes (#1542) (@houko) ### Maintenance - Cover claude code skip permissions args (#1364) (@TechWizard9999) - Fix 16 Dependabot security alerts (#1438) (@SenZhangAI) - Translate all Chinese comments to English (#1509) (@houko) ### Other - Feature/opentel (#1516) (@Chukwuebuka-2003) - Feature/fix gitignore (#1517) (@houko)

LibreFang 2026.3.23

2026-03-23T00:00:00Z

### Added - Add pipeline runner agents + IMAP email reader script (#1307) (@devatsecure) - Add ChatGPT device auth flow (#1332) (@poruru-code) - Add Qwen International and US provider endpoints (#1370) (@houko) - Add custom log directory config (#1379) (@houko) - Enrich ClassifiedError with provider/model context (#1380) (@houko) - Add rustfmt.toml for consistent code formatting (#1381) (@houko) - Display version and git hash in startup logs (#1382) (@houko) - Add unfurl_links config option for Slack channel (#1383) (@houko) - Add DeepInfra as LLM provider (#1384) (@houko) - Add configurable embedding dimensions (#1386) (@houko) - Add config validation with tolerant mode (#1387) (@houko) - Add Azure OpenAI provider support (#1388) (@houko) - Add force_flat_replies config for Slack channels (#1390) (@houko) - Add fts_only mode for memory indexing without embedding (#1391) (@houko) - Add global workspace directory for cross-session persistence (#1392) (@houko) - Add mention_patterns config for Discord channels (#1394) (@houko) - Add WorkflowTemplate types and in-memory registry (#1395) (@houko) - Add configurable session reset prompt (#1396) (@houko) - Add per-agent plugin scoping with allowed_plugins (#1399) (@houko) - Add /reboot slash command for graceful context reset (#1401) (@houko) - Support arbitrary config keys in skill entries (#1402) (@houko) - Add Homebrew Cask CI sync and improve Formula generation (#1404) (@houko) - Comprehensive React dashboard UI/UX overhaul (#1419) (@houko) - Add refresh param to bypass worker cache for migration (#1426) (@houko) - Add Japanese dashboard localization (#1427) (@poruru-code) - Add a new Librefang promotional SVG banner and update the corre… (#1429) (@houko) - Just api starts dashboard dev server alongside API (#1434) (@houko) - Implement depends_on DAG execution for workflow steps (#1440) (@houko) - Add workflow template API endpoints (#1442) (@houko) - Wire thinking model configuration into agent loop (#1443) (@houko) - Mobile responsive + PWA + login + skill output persistence (#1445) (@houko) - Implement session context injection with multiple sources (#1448) (@houko) - Save existing workflow as reusable template (#1449) (@houko) - Add Shell/Bash skill runtime (#1450) (@houko) - Add push messaging API for agents to send to channels (#1451) (@houko) - Add /btw ephemeral side question command (#1452) (@houko) - Add structured output (JSON/JSON Schema) for agents (#1453) (@houko) - Add session export/import for context hibernation (#1454) (@houko) - Configurable heartbeat timeout and pruning per agent (#1455) (@houko) - Cross-session wake via target_agent on triggers (#1456) (@houko) - Add interactive message payloads for Telegram and Slack (#1457) (@houko) - Add PII privacy controls with pseudonymization and redaction (#1458) (@houko) - Tool-level authorization with per-sender and channel-specific policies (#1459) (@houko) - Subagent context inheritance in workflow steps (#1460) (@houko) - Lazy-load LLM driver cache for improved runtime performance (#1461) (@houko) - Add Amazon Bedrock embedding driver with SigV4 signing (#1462) (@houko) - FTS5 full-text session search with API endpoint (#1463) (@houko) - Message injection between tool calls (mid-turn interrupt) (#1464) (@houko) - Render LaTeX in chat (#1467) (@TechWizard9999) - Automatic memory chunking for long documents (#1468) (@houko) - Input sanitizer for prompt injection detection (#1469) (@houko) - Add Android (aarch64) cross-compilation for Termux users (#1470) (@houko) - Time-based memory decay for hierarchical memory management (#1471) (@houko) - File-based input inbox for async external commands (#1472) (@houko) - Interactive approval dialog in dashboard chat and channel events (#1474) (@houko) - Telegram thread-based agent routing (#1475) (@houko) - Pause/resume, busy guard, AgentManifest composition (#1482) (@houko) - Add librefang-testing crate with mock infrastructure (#1483) (@houko) - Show GitHub compare link before version confirmation (#1488) (@houko) - Integrate Skillhub marketplace as second skill source (#1504) (@houko) - Add WeChat personal account adapter via iLink protocol (#1506) (@houko) - Comprehensive build automation CLI with 31 subcommands (#1511) (@houko) - Enhance Hand system with i18n, pause/resume, and dashboard overhaul (#1515) (@houko) - Enable by default, add Grafana, auto-start with Docker (#1520) (@houko) - Multi-agent hand architecture (#1521) (@houko) - Add regex group trigger patterns (#1529) (@TechWizard9999) - Generic media generation drivers (image, TTS, video, music) (#1532) (@houko) - Extend Prometheus metrics and add Grafana dashboards (#1533) (@houko) - Add LTS version support (#1535) (@houko) ### Fixed - Handle paginated /api/agents response (#1233) (@f-liva) - Preserve caption on Telegram voice messages (#1249) (@f-liva) - Correct language toggle logic in navigation sidebar (#1349) (@danilopopeye) - Escape < in MDX comparison table to fix build (#1350) (@houko) - Escape < in MDX troubleshooting page (#1351) (@houko) - Resolve compilation errors breaking CI clippy check (#1353) (@houko) - Clean stale registry dir before clone to prevent CI race condition (#1356) (@houko) - Handle re-release in release.sh when no files changed (#1360) (@houko) - Register aliases for custom models (#1366) (@TechWizard9999) - Knowledge_query JOIN matches entities by name or ID (#1369) (@houko) - Browser hand connection failure on Windows (#1371) (@houko) - Infinite retry guard, dead branch cleanup, body size limit (#1372) (@houko) - Workflow editor save handles nested mode/error_mode from frontend (#1373) (@houko) - Scope knowledge JOIN by agent_id and add entities.name index (#1374) (@houko) - Replace fragile cmd.len() < 50 heuristic in LoopGuard poll detection (#1378) (@houko) - Fix sidebar navigation, broken links, and i18n issues (#1385) (@houko) - Comprehensive website polish and bug fixes (#1389) (@houko) - Accept [hand] wrapper in HAND.toml format (#1393) (@houko) - Fix OG image, brand naming, PWA manifest, and missing i18n keys (#1397) (@houko) - Improve Qwen Code CLI path detection (#1398) (@houko) - Respect provider field when routing custom models (#1400) (@houko) - Remove empty sections overrides and fix mobile nav indicators (#1406) (@houko) - Correct Docker compose port binding for admin interface (#944) (#1407) (@houko) - Allow hyphens in MCP server names (#947) (#1408) (@houko) - Resolve GitHub stats zeros and optimize KV operations (#1409) (@houko) - Load .env files in desktop app (#1410) (@houko) - Prevent streaming interrupts during multi-tool sequences (#1411) (@houko) - Resolve skill file paths for installed skill execution (#1412) (@houko) - Detect and retry when LLM skips tool execution for action requests (#1413) (@houko) - Cache workspace and skill metadata to reduce per-message overhead (#1414) (@houko) - Stop agent loop on tool execution failure (#948) (#1415) (@houko) - Replace processed images with text placeholders in session history (#911) (#1416) (@houko) - Complete ChatGPT Responses driver streaming/tool/reasoning mapping (#1405) (#1421) (@houko) - Migrate old KV keys to history blob and handle sparse chart data (#1422) (@houko) - Complete dashboard i18n coverage for goals and analytics (#1423) (@poruru-code) - Correct provider counts, model numbers, and free tier status (#1424) (@houko) - Update Hands count to 14 and add deploy/registry links (#1428) (@houko) - Release.sh grep compatibility on macOS (#1431) (@houko) - Correct Cloudflare Pages _redirects SPA fallback format (#1432) (@houko) - Release.sh — macOS grep compat + full diff link (#1433) (@houko) - Generate anchor IDs for h3 headings and preserve TOML-style names (#1435) (@houko) - Use 2-digit year in Tauri version for WiX MSI compatibility (#1439) (@houko) - Harden workflow permissions and catalog path validation (#1444) (@SenZhangAI) - Stabilize nodeTypes to fix workflow builder editing (#1447) (@houko) - Harden reconnect and request handling (#1465) (@TechWizard9999) - CI shell injection, clippy warnings, init config, and review findings (#1473) (@houko) - Validate tool_use.input as dict in Anthropic and OpenAI drivers (#1476) (@houko) - Replace plaintext password with Argon2id hashing (#1477) (@houko) - Replace git-based registry sync with HTTP tarball download (#1479) (@houko) - Hand registry race condition, state persistence, and optional requirements (#1481) (@houko) - Resolve clippy errors blocking all PRs (#1486) (@houko) - Consolidate confirmations into single final prompt (#1491) (@houko) - Align chat websocket contract (#1498) (@poruru-code) - Exempt non-autonomous agents from timeout check (#1499) (@houko) - Stamp last_active before LLM call (#1500) (@houko) - Reset last_active on agent restore (#1501) (@houko) - Resolve clippy and compilation errors from merged PRs (#1502) (@houko) - Use tokio::test for callback query tests (#1503) (@houko) - Resolve compilation and clippy errors from recent merges (#1507) (@houko) - Update tool fallback assertions for capability enforcement (#1508) (@houko) - Follow up merged PR regressions (#1514) (@houko) - Use endpoint discovery API for Feishu WebSocket connection (#1518) (@houko) - Gitignore, channel logging, and xtask Windows CI (#1519) (@houko) - Preserve coordinator role and role-bound trigger migration (#1523) (@houko) - Restore --release flag in Dockerfile build (#1524) (@houko) - Eliminate username enumeration timing side-channel (#1525) (@houko) - Replace deterministic session token with random generation (#1526) (@houko) - Prevent path traversal in skill script execution (#1527) (@houko) - Make init_prometheus idempotent for parallel test safety (#1528) (@houko) - Multi-agent parsing compat + registry sync version update (#1530) (@houko) - Gate unix-only test behind #[cfg(unix)] (#1534) (@houko) - Release tool compares against latest tag including prereleases (#1547) (@houko) - Release tool retries commit after formatter hook (#1548) (@houko) ### Changed - Switch to CalVer (YYYY.M.DDHH) (#1375) (@houko) - Add facade getters and migrate API routes (#1478) (@houko) - Modularize route registration into per-domain routers (#1484) (@houko) - Split monolithic config.rs (5566 LOC) into modular sub-modules (#1485) (@houko) - Registry as catalog, pre-install core content only (#1537) (@houko) - Unified workspaces layout + hand/agent isolation + routing fixes (#1542) (@houko) ### Documentation - Comprehensive review — fix errors, update numbers, add missing sections (#1368) (@houko) ### Maintenance - Lock api status version regression (#1363) (@TechWizard9999) - Cover claude code skip permissions args (#1364) (@TechWizard9999) - Cover hand reactivation runtime profile (#1365) (@TechWizard9999) - Cover local model default override routing (#1367) (@TechWizard9999) - Auto-update PR branches on main push (#1417) (@houko) - Add GitHub Stats Worker to deploy workflow (#1420) (@houko) - Remove deploy worker job-level if conditions that fail on squash merges (#1425) (@houko) - Fix 16 Dependabot security alerts (#1438) (@SenZhangAI) - Translate all Chinese comments to English (#1509) (@houko) ### Other - Feature/opentel (#1516) (@Chukwuebuka-2003) - Feature/fix gitignore (#1517) (@houko)

LibreFang 2026.3.22

2026-03-22T00:00:00Z

LibreFang 2026.3.21

2026-03-21T00:00:00Z